Design Of The Defects Detection System For Magnets Surface Based On Machine Vision

Magnetic tile is an important part of the permanent magnet motor, so its surface defect which detected in the production process is a very important part. On the other hand, artificial visual inspection is commonly used by magnetic tile manufacturers at present, it affected the production efficiency and product quality seriously. Therefore, this paper proposes a magnetic tile surface detecting method based on machine vision, in order to achieving a real-time, non-contact, high-efficiency, and high-precision effect.. The reasonable use of this method will have greatly significant for the automatic production enterprises. Specific are summarized as follows:Firstly, a curved direct-type light source is designed pertinently on the basis of analyzing defect types and characteristics of magnetic tile surface, the image processing platform, cameras and lenses is selected as to obtaining high-quality magnetic tile surface image, and image filtering. For the uneven brightness problem of the magnetic tile surface, this paper presents an improved homomorphic filtering algorithm, enhancing image contrast by morphological top-hat and bottom hat transformation, through comparing results of three commonly uneven brightness correction algorithm. The method can balance image brightness effectively, highlighting the defect area to a certain extent.Secondly, combining with image block theory, an improved variance image block feature described by mean, entropy and gray feature quantity is proposed. By comparing the amount of gray feature to determine whether the magnetic tile surface is defected quickly. For texture case, with the method of segmentation based on visual attention mechanism defect image block, and select from the proposed visual features of the method, the integration features significantly improve the visual angle of diagram significant computational model programs.Once again, through a variety of image features analysis, select gray feature including improved mean entropy variance, and seven Hu invariant moments got from binarized images as the basis for defect classification. Then, random forest classifier is used to detect defect image block, identify the category of defect.Finally, based on the above analysis, system software structures, algorithms and software platform are tested to verify the reliability of the platform stability and algorithms.The proposed algorithm can effectively detect magnetic tile surface defects, with better real-time performance and reliability, and lay a solid foundation for the subsequent on-line defect detection field environment.